Circulating and Intratumoral Immune Determinants of Response to Atezolizumab plus Bevacizumab in Patients with Variant Histology or Sarcomatoid Renal Cell Carcinoma

The most common form of renal cell carcinoma (RCC) is clear cell RCC (ccRCC), which accounts for 75% to 80% of the approximately 75,000 cases diagnosed each year in the United States (1). Variant histology RCC [commonly referred to as non–clear cell RCC (nccRCC)], which includes subtypes with distinct biological immunopathologic features (e.g., chromophobe, papillary, translocation) and is associated with adverse outcomes, represents 20% to 25% of RCC (2). In addition, any RCC histology may be associated with sarcomatoid differentiation, which typically confers a more aggressive clinical behavior and worse outcomes (3). Although immune checkpoint inhibition (ICI) has shifted the treatment paradigm of RCC, most large clinical trials have included only patients with ccRCC (4–6). Thus, there is an unmet need to understand the clinical and biological features that determine the therapeutic response to ICIs in patients with variant histology RCC or RCC with sarcomatoid features (7–11).

The combination of atezolizumab, an mAb specific for programmed death-ligand 1 (PD-L1), and bevacizumab, an mAb specific for vascular endothelial growth factor A (VEGF-A), has clinical activity in patients with ccRCC, particularly in specific molecular subtypes (12, 13). In addition, among patients with variant histology RCC or RCC with sarcomatoid differentiation, atezolizumab plus bevacizumab had an acceptable safety profile and clear clinical activity, particularly in tumors positive for PD-L1 expression or sarcomatoid differentiation (14). Moving forward, it will be critical to determine the biological factors that determine therapeutic response, with the goal of informing treatment selection and sequencing.

Although the International Metastatic RCC Database Consortium (IMDC) prognostic model continues to perform well in variant histology RCC, the circulating and intratumoral factors that determine response to ICI in these tumor types remain largely unknown (15). Herein, we performed a blood- and tissue-based biomarker analysis of a multicenter phase II clinical trial of atezolizumab and bevacizumab for patients with metastatic RCC (mRCC) of variant histology or any histology with sarcomatoid features (14). In this phase II clinical trial of atezolizumab and bevacizumab that included 60 patients with variant histology mRCC or any mRCC histology with ≥20% sarcomatoid differentiation, the objective response rate was 33% in the overall study population, 50% in patients with ccRCC with sarcomatoid differentiation and 26% in patients with variant histology RCC (14). In this study, we defined baseline circulating inflammatory (IL1, IL6, IL13, MCP-1, and MIP-1β) and angiogenic soluble (VEGF-A) factors associated with clinical risk and therapeutic outcome. We also sought to assess the association between the presence of intratumoral terminally exhausted CD8⁺ T cells and clinical outcomes. This study provides insights into the biological factors that impact response to ICI and anti-angiogenic agents in variant histology RCC and supports future validation studies of circulating and intratumoral biomarkers in these diseases.

This study used patient data and biospecimens [blood (N = 60) and tumor (N = 38)] from a prospective phase-II clinical trial of the anti–PD-L1 atezolizumab and the anti–VEGF-A bevacizumab in patients with mRCC with variant histology and/or sarcomatoid differentiation (NCT02724878; ref. 14). This study was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants before their participation in the study. Blood samples were collected at baseline, before therapy (C1D1: cycle one day one), and on-treatment, following two cycles of therapy (C3D1: cycle three day one). Plasma was assayed for the levels of soluble factors (e.g., cytokines and growth factors), and peripheral blood mononuclear cells (PBMC) were immunophenotyped using flow cytometry. For Dana-Farber/Harvard Cancer Center (DF/HCC) participating sites, 30 mL of blood were collected as per the Laboratory Manual and couriered within 6 hours of collection to The Center of Immunooncology Laboratory. For non-DF/HCC participating sites, 30 mL of blood were collected and stored as per the Laboratory Manual and batch shipped at the end of study to The Center of Immunooncology Laboratory. In addition, formalin-fixed, paraffin-embedded (FFPE) archival or pretreatment study tumor biopsies were analyzed by IHC and immunofluorescence for enrolled subjects with available biospecimen. Archived tumor tissue was collected prestudy as per the Laboratory Manual and stored at ambient temperature until the end of study at which time the specimens were shipped to the Signoretti Laboratory. Fresh-tumor biopsy was conducted prestudy as per the Laboratory Manual. Frozen specimens were stored at −80°C and specimens fixed in formalin were stored at 4°C until shipping. At the end of study, the tumor specimens were shipped to the Signoretti Laboratory.

Baseline demographics, clinical characteristics, and treatment outcomes, including progression-free survival (PFS), overall survival (OS), and clinical benefit (CB; ref. 16), were collected and analyzed from the trial database. PFS was defined as the time from treatment initiation to radiographic progression or death from any cause or censored at the date of last disease assessment. OS was calculated from treatment initiation to the date of death from any cause or censored at the date of last follow-up. Objective response was assessed by the clinical investigator using RECIST version 1.1 every 6 weeks during the first 24 weeks and then every 12 weeks after treatment initiation until radiographic progression occurred, intolerable toxicity, or withdrawal of consent. CB was defined (16) by the sum of objective responses (complete or partial) and stable disease where patients had tumor shrinkage and PFS ≥6 months. Patients with progressive disease and PFS <3 months were classified as having no CB (NCB). All other patients were classified as having intermediate CB (ICB; ref. 17). This study was approved by each participating institution's institutional review board.

The Luminex fluorescent bead array platform (Bio-Plex MAGPIX) was used to quantify multiple soluble biomarkers simultaneously in human plasma at baseline and on-treatment (Bio-Rad, Inc.). The samples were analyzed in duplicates, and the fluorescent values were extrapolated to a standard curve for quantification. The following markers were analyzed: EGF, IFNg, IFNa2, TNFa, TNFb, IL1b, IL1a, IL1RA, IL2, sIL2R, IL3, IL4, IL5, IL6, IL7, IL8, IL9, IL10, IL12, IL13, IL15, IL17, MIP-1a, MIP-1β, MCP-1, MCP-3, CXCL6, CXCL5, IP-10, GROb, VEGF-A, FGF2, Ang-1, Ang-2, Eotaxin, TGFa, G-CSF, Flt-3L, Fractalkine, and sCD40 L and sPDL-1 (R&D systems).

For flow cytometry analysis, PBMCs were thawed in a 37°C water bath and transferred to a 15 mL conical tube containing warm DMEM (GIBCO, 11965–092). Cells were pelleted at 1,800 RPM at 4°C (Sorvall Legend XTR Centrifuge, Thermo Fisher Scientific) for 5 minutes and washed with PBS (Gibco). To discriminate between living and dead cells, PBMCs were incubated with Near Infrared Fixable Viability Dye (NIR, BioLegend; Invitrogen LIVE/DEAD Fixable Near-IR Dead Cell Stain Kit, for 633 or 635-nm excitation; L10119) in PBS for 15 minutes on ice and then washed once with 1-mL PBS per the manufacturer's protocol. PBMCs were next incubated on ice with Fc block (Human TruStain FcX BioLegend, 422302) in FACS buffer (PBS/2.5% FBS) for 20 minutes, washed once, and divided into single tubes (approximately 500,000 cells per tube). Next, cells were incubated on ice with the antibodies listed in Supplementary Table S1 in PBS/2.5% FBS. Cells were washed twice with FACS buffer and kept in the dark on ice until flow cytometry analysis (no more than 30 minutes after the last wash). Data were acquired using a Fortessa analyzer (BD Biosciences), and FCS files were analyzed using FlowJo software 10.7.1 (Treestar, Inc.). The gating strategy that was adopted is presented in Supplementary Figs. S1–S3.

Tumor tissue samples (N = 38) were collected before the initiation of therapy and analyzed using IHC and multiplex immunofluorescence (MIF).

The assay was performed on a Bond RX Autostainer (Leica Biosystems) using the Opal multiplex IHC system (PerkinElmer/Akoya Biosciences Cat# NEL871001KT) as described in detail previously (18). A previously optimized panel of antibodies specific for CD8 (Agilent, M710301–2, RRID:AB_2075537), PD-1 (clone EH33, a gift from Dr. Gordan Freeman, Dana-Farber Cancer Institute), TIM3 (R&D Systems, AF2365, RRID:AB_355235), and LAG3 (LifeSpan Biosciences, LS-B2237–50, RRID:AB_1650048) was used (Supplementary Table S2). In brief, whole-slide multispectral images were acquired at ×10 magnifications using Vectra 3.0 (PerkinElmer/Akoya Biosciences). Next, they were visually inspected using Phenochart 1.0.9 (PerkinElmer/Akoya Biosciences, RRID:SCR_019160) to select at least five intratumoral CD8⁺ T cell–enriched regions of interest (ROI) per case where each ROI had an area = 669 × 500 μm. The number of hotspot ROIs was determined so that a minimum of 100 CD8⁺ T cells would be analyzed. ROIs were scanned at ×20 magnifications, imported into InForm v2.2.0 (PerkinElmer, RRID:SCR_019161) and deconvoluted using a multispectral library built with single-stain slides. Image analysis was performed using HALO v2.1.1637.18 (Indica Labs, RRID:SCR_018350) and the Indica Labs High-Plex FL v2.0 module. A unique algorithm was created for each case, and its accuracy was validated through visual inspection by two pathologists (T. Denize and M. Ficial) with extensive experience in image analysis.

Tumor sections had been previously stained with a validated antibody against PD-L1 (05.9A11 mouse mAb, 1:125, 13 μg/mL; a gift from Dr. Gordon Freeman, and commercially available through Cell Signaling Technology) with BOND RX autostainer (Leica Biosystems) using the BOND Polymer Refine Detection Kit (DS9800; Leica Biosystems) and BOND Polymer Refine Red Detection Kit (DS9390; Leica Biosystems), as previously described (14). The PD-L1 score was calculated by dividing the number of PD-L1⁺ tumor cells by the total number of tumor cells. A cutoff value of 1% was used to determine PD-L1 positivity.

To characterize the immune landscape of patients with mRCC with variant histology and/or sarcomatoid differentiation, the relationships between biomarker variables were analyzed using Pearson correlations and hierarchical clustering using Euclidean distance. All biomarker variables were Z-score normalized. We calculated an inflammatory module, the systemic inflammation score (SIS), which corresponds to the average score of the five highly correlated inflammatory cytokines: MIP-1β, IL1, MCP-1, IL6, and IL13. To evaluate the association of SIS with IMDC risk groups and CB groups, pairwise Wilcoxon signed-rank tests were applied. The relationships of SIS with PFS and OS were assessed by log-rank tests and univariate Cox regression models. PFS and OS were estimated using the Kaplan–Meier methodology. Biomarkers were dichotomized on the basis of the median for all measures and the 25th percentile for SIS. Values were consistent across duplicates. To account for the different proportions of cells obtained across gating strategies, we calculated an average score for each cell type. All tests were two-sided with a significance threshold of 0.05 and R (version 4.1.1) was used to carry out the above analyses. We conducted all our analyses in the entire cohort and in different subsets according to sarcomatoid status and histology.

The data generated in this study are available in the in this article and its Supplementary Data Files or upon request from the corresponding author.

We determined circulating plasma-derived cytokine signatures and peripheral immune-cell profiles of all 60 patients at baseline and after two cycles of therapy to explore determinants of response to treatment and survival outcomes. In addition, we used IHC and MIF on baseline FFPE tumor biospecimens of 38 patients to identify intratumoral T-cell phenotypes associated with response to therapy (Fig. 1; Supplementary Table S3).

To identify plasma cytokine and soluble factor signatures associated with response to therapy, we investigated the distribution of each cytokine in the whole cohort and in sarcomatoid status–defined subgroups (Supplementary Fig. S4). Apart from two angiogenic cytokines (VEGF-A and Angiopoietin 2), which were more abundant in patients with nonsarcomatoid tumors, circulating cytokines had similar distributions between sarcomatoid and nonsarcomatoid subgroups. We performed pairwise correlations between all quantified plasma proteins at baseline. Hierarchical clustering of this correlation matrix revealed an “inflammatory module” consisting of highly correlated levels of five critical inflammatory cytokines (IL1, IL6, IL13, MIP-1β, and MCP-1; Fig. 2A). Inflammatory proteins also strongly correlated in the sarcomatoid (IL1, IL6, MCP-1, IL13, MIP-1β; Supplementary Fig. S5A) and nonsarcomatoid (MIP-1β and IL6; Supplementary Fig. S5B) subgroups. CD40L, GROb, and EGF also appeared to be highly correlated in the overall cohort (Fig. 2A), and in the sarcomatoid (Supplementary Fig. S5A) and nonsarcomatoid (Supplementary Fig. S5B) subgroups. However, we observed no clear association between CD40L, GROb, and EGF (individually or as a module) with PFS (Supplementary Fig. S5C). By contrast, when we individually examined the five inflammatory cytokines (IL1, IL6, IL13, MIP-1β, and MCP-1), elevated levels of four of these five cytokines corresponded with lower PFS (Fig. 2B). Cox regression analyses uncovered negative associations between PFS and plasma levels of IL1 (HR, 1.71; P = 0.004), IL6 (HR, 1.70; P = 0.012), IL13 (HR, 1.64; P = 0.014), and MIP-1β (HR, 2.05; P = 0.007). In addition, higher expression levels of IL1 (HR, 1.38; P = 0.02), MIP-1β (HR, 1.52; P = 0.003), IL6 (HR, 1.79; P = 0.002), and IL13 (HR, 1.48; P = 0.009) were individually associated with worse OS (Supplementary Fig. S5D). We subsequently calculated a SIS based on the mean level of these five correlated inflammatory cytokines. Elevated SIS was associated with worse PFS (Cox regression: HR, 2.24; P < 0.001; Fig. 2C) and OS (HR, 1.76; P = 0.003 Fig. 2C). The same trend appeared in the sarcomatoid (HR, 4.31; P = 0.006) and nonsarcomatoid (HR, 1.6; P = 0.602) subgroups as well as in CC+sarcomatoid (HR, 3.06; P = 0.07), chromophobe (HR, 5.07; P = 0.057), collecting duct (HR, 343391; P = 0.208), and other (HR, 12.2; P = 0.127) histologies (Supplementary Fig. S5E). Similar results were found for OS in these subgroups as well as in translocation and unclassified histologies (Supplementary Fig. S5E). In the overall population, the SIS (continuous variable) was associated with worse PFS and OS. When dichotomized into “high” and “low” groups, higher SIS was associated with worse survival at numerous thresholds (Fig. 2C; Supplementary Fig. S5F). Patients with higher SIS had improved PFS (log-rank test, P = 0.028; Fig. 2D) and OS (log-rank test, P = 0.036; Supplementary Fig. S5G). Even after controlling for the IMDC score in a multivariate analysis, we observed a trend of association between high SIS and worse PFS (HR, 1.43; P = 0.099) and OS (HR, 1.30; P = 0.093).

We observed that an elevated SIS score was also associated with a lack of CB (Wilcoxon test, NCB vs. CB/ICB, P = 0.033, Supplementary Fig. S6A). When comparing individual subgroups, we found that an elevated SIS score was associated with a lack of CB as compared with intermediate ICB (Wilcoxon test, NCB vs. ICB, P = 0.044; Fig. 2E). A similar (nonsignificant) trend was observed for CB versus NCB (Wilcoxon test, P = 0.058; Fig. 2E). In subgroup analyses, higher SIS was associated with resistance to therapy in patients with sarcomatoid differentiation and increased baseline (Supplementary Fig. S6B); however, those without sarcomatoid component did not show any association between SIS and clinical response to treatment (Supplementary Fig. S6C). The same trends were consistent across histologies (Supplementary Fig. S6D). SIS also differed between IMDC risk groups, with the poor-risk group having significantly higher inflammation scores than the intermediate-risk (Wilcoxon test, P = 0.03) and favorable-risk (Wilcoxon test, P = 0.01; Fig. 2F) groups. In a subgroup analysis by sarcomatoid versus nonsarcomatoid tumors, the same pattern was observed, with patients in the IMDC poor-risk group exhibiting higher SIS (Supplementary Fig. S7A and S7B). This result was consistent across different histologies (Supplementary Fig. S7C).

VEGF-A levels at baseline were higher in patients with therapy-resistant disease (NCB) compared with patients with responsive disease (CB; Wilcoxon test, P = 0.027) or ICB (Wilcoxon test, P = 0.01; Fig. 3A). More broadly, VEGF-A levels were higher in resistant disease (NCB) compared with nonresistant disease (CB+ICB; Wilcoxon test, P = 0.0137; Supplementary Fig. S8A). In the sarcomatoid subgroup, patients with higher baseline VEGF-A levels exhibited resistance to treatment when compared with patients who benefited from atezolizumab+bevacizumab (P = 0.04; Supplementary Fig. S8B). A similar (nonsignificant) trend was observed in the nonsarcomatoid subset (P = 0.6; Supplementary Fig. S8C). The association of baseline VEGF-A levels and CB in individual histologies were overall concordant (Supplementary Fig. S8D). Higher baseline VEGF-A levels (as a continuous variable, and dichotomized at the median) were associated with lower PFS (Cox regression, HR, 2.017; P = 0.001; Supplementary Fig. S8E; log-rank test, P = 0.021; Fig. 3B) and OS (Cox regression, HR, 1.75; P < 0.001; Supplementary Fig. S8E; log-rank test; P = 0.0031; Fig. 3C). In addition, a higher baseline VEGF-A level tended to be associated with shorter PFS independently of sarcomatoid status (sarcomatoid: HR, 2.02; P < 0.001; nonsarcomatoid: HR, 1.06; P = 0.876) and in all histologies except papillary and translocation RCC (Supplementary Fig. S8E).

We defined the dynamic evolution of circulating VEGF-A as Δ (VEGF-A) = [VEGF-A(C3D1)] − [VEGF-A(C1D1)]/VEGF-A(C1D1), and dichotomized it using the median. In contrast with the pattern observed before treatment, those patients who had a greater increase in plasma VEGF-A [high Δ (VEGF-A)] on-treatment (C3D1) compared with baseline (C1D1), had overall improved clinical outcomes than patients with more modest increases or decreases in VEGF-A [low Δ(VEGF-A; Fig. 3D–F)]. Patients with therapy-resistant disease (NCB) had decreases or only modest increases in plasma VEGF-A as compared with patients who responded to treatment (CB; Wilcoxon test, P = 0.012), or to those with ICB (Wilcoxon test, P = 0.013; Fig. 3D), and more broadly to those with nonresistant disease (CB+ICB; P = 0.0058; Supplementary Fig. S9A) in the overall cohort. The same trend was observed in different subgroups according to sarcomatoid status or individual histology (Supplementary Fig. S9B–S9D). Patients with larger increases in plasma VEGF-A levels during therapy also tended to have improved PFS [Cox regression, HR, 0.78; P = 0.33; Supplementary Fig. S9E; log-rank test, P = 0.059; Fig. 3E (Kaplan–Meier curve starting at C1D1); P = 0.12; Supplementary Fig. S9F (Kaplan–Meier curve starting at C3D1)] and OS [Cox regression, HR, 0.12; P = 0.01; Supplementary Fig. S9E; log-rank test, P = 0.0058; Fig. 3F (Kaplan–Meier curve starting at C1D1); P = 0.0014; Supplementary Fig. S9G (Kaplan–Meier curve starting at C3D1)].

To better understand how the circulating immune landscape impacts response to therapy in variant histology RCC, we used known lineage markers to analyze the baseline peripheral blood samples by high-dimensional flow cytometry (Fig. 4). Across all samples, we found no large modules of correlated immune cell types at baseline (Pearson correlations, Fig. 4A). We did, however, find areas of moderate correlation among CD4⁺PD-1⁺ and CD4⁺ T-cell effector memory populations (TEM, r = 0.64; P < 0.0001; and TEMRA, r = 0.75; P < 0.0001). In addition, CD4⁺PD-L1⁺ T-cell levels correlated with CD8⁺PD-L1⁺ T-cell levels (r = 0.71; P < 0.0001; Fig. 4A).

To further understand the importance of PD-L1⁺ T cells, we investigated their relative changes at C3D1 compared with C1D1. The proportion of CD8⁺PD-L1⁺ T cells decreased with treatment in all patients (Fig. 4B). However, patients with therapeutic response (CB) had a smaller decrease in the proportion of CD8⁺PD-L1⁺ T cells than patients with therapeutic resistance (NCB) with atezolizumab plus bevacizumab treatment (P = 0.06; Fig. 4B). The larger decrease in CD8⁺PD-L1⁺ T cells appeared to specifically be associated with resistance (Wilcoxon test, NCB vs. CB+ICB, P = 0.09; Supplementary Fig. S10A). However, a similar trend was not clearly observed with the relative change in CD4⁺ PD-L1⁺ T cells on-treatment (P = 0.22; Supplementary Fig. S10B). Furthermore, we did not find any associations between baseline PD-L1⁺ T cells and outcomes in different subsets (Supplementary Fig. S10C–S10E).

Differences in the relative changes in PD-L1⁺ T cells on-treatment were also associated with altered PFS and OS. A larger decrease of PD-L1⁺ T cells on-treatment was associated with worse PFS for both CD4⁺PD-L1⁺ T cells (HR, 1.61; P = 0.016; Fig. 4C) and CD8⁺PD-L1⁺ T cells (HR, 1.69; P = 0.009; Fig. 4C). The same trend was observed for OS for both CD4⁺PD-L1⁺ T cells (HR, 1.49; P = 0.12) and CD8⁺PD-L1⁺ T cells (HR, 1.75; P = 0.025; Supplementary Fig. S11A and S11B). For patients with higher (>median) on-treatment decrease in CD8⁺PD-L1⁺ T cells, the median PFS was 5.5 months and median OS was 21.3 months. In contrast, among patients with lower (≤median) on-treatment decreases in CD8⁺PD-L1⁺ T cells, the median PFS and median OS were not reached [PFS, P = 0.02; Fig. 4D (Kaplan–Meier curve starting at C1D1); P = 0.056; Supplementary Fig. S11C (Kaplan–Meier curve starting at C3D1); OS, P = 0.036; Supplementary Fig. S11D and S11E]. Among both sarcomatoid/nonsarcomatoid groups and different histologies, a higher decrease in PD-L1⁺ circulating T cells remained associated with worse PFS (Supplementary Fig. S11A and S11B).

Because the tumor microenvironment also plays a crucial role in determining the response ICI, we assessed PD-L1 expression on tumor cells by IHC. In addition, we performed detailed spatial phenotyping of T cells, investigating the expression of markers of antigen experience (PD-1) and terminal exhaustion (TIM-3 and LAG-3) on CD8⁺ tumor-infiltrating T cells using MIF.

The average PD-L1 expression was 19.6% of tumor cells across sarcomatoid tumors and 27.2% across nonsarcomatoid tumors (Fig. 5A and B). There were no differences in PD-L1 expression between sarcomatoid and nonsarcomatoid tumor samples (Wilcoxon test, P = 0.77). There were no associations between the various histologies and their tumor cell PD-L1 expression.

There was no clear association between RCC histology and the proportion of antigen-experienced (PD-1⁺ but TIM-3⁻ and LAG-3⁻) or terminally exhausted (PD-1⁺ with TIM-3⁺ or LAG-3⁺) CD8⁺ T cells, although the analysis was restricted by a limited number of available tumor samples. There was no significant difference in the density of terminally exhausted CD8⁺ T cells between clear-cell sarcomatoid (mean = 151.2 cell/mm²), non–clear cell sarcomatoid (mean = 20.3 cell/mm²) and nonsarcomatoid (mean = 76.5 cell/mm²) tumors (P > 0.5; Fig. 5A). In addition, no significant differences were observed in the density of antigen-experienced CD8⁺ T cells between clear-cell sarcomatoid (mean = 23.7 cell/mm²), non–clear cell sarcomatoid (mean = 114.1 cell/mm²), and nonsarcomatoid (mean = 97.4 cell/mm²) tumors (P > 0.5; Fig. 5A). Of note, the two tumor samples with papillary histology had a relatively low density of terminally exhausted CD8⁺ T cells (3.6 and 7.2 cell/mm²; Fig. 5B).

Across all histologies, there was a clear association between a higher percentage of intratumoral terminally exhausted CD8⁺ T cells and worse PFS. Patients with high (>median) percentages of terminally exhausted CD8⁺ T cells had a median PFS of only 6.9 months, whereas those with low percentages of terminally exhausted CD8⁺ T cells did not reach the median PFS (log-rank test, P = 0.028, Fig. 5C). A higher percentage of intratumoral terminally exhausted CD8⁺ T cells was associated with worse outcomes in patients with nccRCC without sarcomatoid component (P = 0.032, Supplementary Fig. S12B), and a similar (nonsignificant) effect was seen in sarcomatoid tumors (P = 0.33, Supplementary Fig. S12A). However, there was no significant association between the percentage of terminally exhausted CD8⁺ T cells (log-rank test, P = 0.52, Supplementary Fig. S12C) or CD8⁺PD-1⁺ T cells (Supplementary Fig. S12D) and OS.

This correlative analysis of blood and tumor specimens from the phase II clinical trial of atezolizumab+bevacizumab in variant histology RCC revealed both circulating and intratumoral determinants of clinical outcomes with ICI and antiangiogenic agents. Analysis of soluble factors in plasma revealed a highly coordinated “inflammatory module” of circulating immunomodulatory cytokines (IL1, IL6, IL13, and MIP-1β) associated with adverse clinical outcomes. Beyond static baseline measurements, dynamic changes in circulating angiogenic factors (e.g., VEGF-A) and peripheral immune populations (e.g., CD8⁺PD-L1⁺ T cells) also affected clinical outcomes. Finally, within the tumor, a high percentage of terminally exhausted CD8⁺ T cells in the tumor microenvironment were associated with worse PFS. These results provide novel insights into the biology of variant histology RCC or ccRCC with sarcomatoid features in the immunotherapy era and a reference framework for future biomarker studies. Many of these findings build on prior observations in ccRCC and other tumor types, and it is notable that the heterogeneous underlying biology of different RCC histologies and sarcomatoid differentiation does represent a limitation. Nevertheless, our main findings linking systemic inflammatory cytokines, baseline and dynamic changes in VEGF-A, dynamic changes in circulating T-cell populations, and intratumoral terminally exhausted CD8⁺ T cells with altered clinical outcomes have not been described in variant/sarcomatoid RCC histologies.

We observed that increased baseline levels of circulating proinflammatory cytokines are correlated with poorer clinical outcomes, which is consistent with previous observations (12, 19–22). The IMDC-risk score for metastatic RCC, which has been validated as a reliable predictor of OS in patients with variant histology or “non–clear cell” RCC (15), includes several systemic inflammatory correlates, including neutrophilia, anemia, and thrombocytosis, as negative prognostic factors (23). It is notable that SIS discriminates between response and resistance even in the sarcomatoid group, which is typically characterized by a highly inflammatory state. In the IMmotion150 trial, which enrolled patients with clear-cell mRCC (including tumors with a sarcomatoid component), an elevated myeloid inflammation signature within the tumor was associated with shorter PFS in patients receiving atezolizumab-containing regimens (12). Furthermore, anti-IL1β therapy also remodeled the myeloid compartment and delayed tumor outgrowth in the RENCA RCC model (19), suggesting a potential therapeutic approach. In fact, in the CANTOS trial that was designed to test directly the inflammatory hypothesis of atherothrombosis, patients who received canakinumab, an anti-IL1, showed lower cancer mortality (20, 21). In addition, baseline circulating IL6 has been implicated as a potential negative predictive biomarker of response to ICI in non–small cell lung cancer (NSCLC), and in vitro experiments indicate that IL6 can upregulate PD-L1 via JAK1/Stat3 signaling (22). Other circulating inflammatory cytokines, such as IL8, have also been shown to have a negative predictive value for clinical outcomes in patients treated with ICI (24). Further investigations of the immunosuppressive role of systemic inflammation in RCC with variant histology, as well as corresponding circulating biomarkers, are warranted.

Our observation that increased baseline VEGF-A carries a worse prognosis is established in the literature (25, 26), although the implications of this finding for therapy selection are unclear. For instance, Hegde and colleagues (27) analyzed several clinical trials evaluating bevacizumab for metastatic colorectal cancer, NSCLC, and mRCC. They found that increased baseline plasma VEGF-A levels conferred a worse OS prognosis; however, VEGF-A appeared to be a prognostic, but not a predictive biomarker. On the other hand, a meta-analysis of clinical trials investigating bevacizumab for breast cancer found that bevacizumab improved PFS for patients with high (>median) but not low (≤median) VEGF-A levels; ref. 28).

Our finding that greater increases in VEGF-A on-treatment are associated with a better prognosis also has uncertain implications. Among patients with clear cell or papillary RCC treated with adjuvant sunitinib or sorafenib after nephrectomy, exposure to these VEGFR inhibitors were previously found to result in host-mediated compensatory increase in circulating VEGF-A (29). This suggests that increases in VEGF-A are associated with target engagement by VEGF pathway inhibitors and may explain our observed association between increased VEGF-A on therapy and improved clinical outcomes. A review of biomarkers of response to bevacizumab by Lambrechts and colleagues (30) found that changes in circulating VEGF-A levels during treatment have been inconsistent across five trials for different cancer types (not including RCC). Furthermore, one study that included patients with platinum-resistant recurrent ovarian cancer treated with bevacizumab and gemcitabine showed that rapid on-treatment decrease of VEGF-A levels is associated with worse response, CB rates, PFS, and OS (31). The significant differences in PFS and OS associated with on-treatment VEGF-A changes are striking and should be explored in future biomarker studies.

Upon treatment with atezolizumab (plus bevacizumab) peripheral CD8⁺PD-L1⁺ populations universally decreased after one treatment cycle. Smaller decreases in the CD8⁺PD-L1⁺ population after treatment were associated with CB and longer PFS. Previous studies in melanoma and NSCLC have demonstrated that increased CD8⁺PD-L1⁺ cells are associated with an immunosuppressed tumor microenvironment and worse clinical outcomes, and have resulted in mixed outcomes in response to immunotherapy (32, 33). In gastric and gastro–esophageal junction adenocarcinoma, there was a correlation of PD-L1 expression on tumor and immune cells (i.e., CD8⁺ T cells) with worse PFS and OS, but only that of tumor cells reached statistical significance (34). In cervical cancer, the presence of PD-L1–expressing tumor-infiltrating lymphocytes predicted poor prognosis (34). However, to our knowledge, no studies have been conducted to evaluate the effect of CD8⁺PD-L1⁺ T cells on the tumor microenvironment in variant RCC or ccRCC with sarcomatoid features. Further work is required to understand why a decrease in CD8⁺PD-L1⁺ T cells was associated with improved clinical outcomes when baseline CD8⁺PD-L1⁺ T cells are otherwise correlated with worse clinical outcomes in other tumor types, and whether CD8⁺PD-L1⁺ T cells could be a predictive biomarker for response to immunotherapy in variant histology (and clear cell) RCC.

Although we did not detect any clear correlations in tumor PD-L1 expression and tumor histology, we recognize that identifying such patterns may be limited by the small number of tumors analyzed. Sarcomatoid RCC has been characterized by high PD-L1 tumoral expression compared with nonsarcomatoid RCC (35). In agreement with other studies in ccRCC (36, 37), increased terminally exhausted intratumoral CD8⁺ T cells were associated with worse survival outcomes. To our knowledge, this is the first study to describe antigen-experienced and terminally exhausted CD8⁺ T cells in variant histology RCC. It would have also been interesting to dissect the distribution of the IMmotion151 molecular clusters (13) across this nccRCC cohort.

In conclusion, when examining the immune landscape of patients with RCC of variant histology treated with atezolizumab plus bevacizumab, both the peripheral (circulating) and intratumor immune topography, were associated with treatment-related outcomes. Furthermore, our data demonstrate that a baseline inflammatory state is associated with worse outcomes, whereas a lower baseline VEGF-A concentration and higher increase in VEGF-A levels on treatment correlated with CB and improved survival. In addition, marked decreases in circulating CD8⁺PD-L1⁺ T cells had a negative prognostic value. Finally, patients with high levels of intratumoral terminally exhausted CD8⁺ T cells exhibited shorter PFS. In aggregate, these findings support the further investigation of both peripheral and intratumoral immune biomarkers as part of an integrative approach in mRCC with variant histology or sarcomatoid differentiation.

T. El Zarif reports personal fees from Art tempi outside the submitted work. Z. Bakouny reports personal fees from UpToDate, grants from Genentech/imCORE, and nonfinancial support from Bristol Myers Squibb outside the submitted work. W. Xie reports personal fees from Convergent Therapeutics outside the submitted work. R. Flippot reports nonfinancial support from Bayer, and personal fees from Janssen, MSD, Astellas, Ipsen, and Pfizer outside the submitted work. L. Hirsch reports personal fees from Gilead outside the submitted work. X.X. Wei reports personal fees from Novartis and other support from BMS outside the submitted work. L.C. Harshman reports current employment and stock options from Surface Oncology. U.N. Vaishampayan reports grants and personal fees from Merck, grants and personal fees from BMS, personal fees from Exelixis, personal fees from Sanofi, personal fees from Gilead, Bayer, Seattle Genetics, Alkermes, Novartis, and Pfizer outside the submitted work. D.F. McDermott reports grants from Genentech during the conduct of the study; grants and personal fees from BMS, Merck, and Pfizer outside the submitted work. W. Xu reports personal fees from Exelixis, Jazz Pharmaceuticals, MedNet, Harborside Press, MJH Healthcare Holdings, and Academy for Continued Healthcare Learning outside the submitted work. E.M. Van Allen reports personal fees from Tango Therapeutics, Genome Medical, Genomic Life, Monte Rosa Therapeutics, Manifold Bio, and Illumina, grants and personal fees from Novartis, grants from BMS, personal fees from Enara Bio, grants and personal fees from Janssen, and personal fees from Foaley & Hoag, Riva Therapeutics, grants from Sanofi, and personal fees from Serinus Bio outside the submitted work; as well as reports a patent for Institutional patents filed on chromatin mutations and immunotherapy response, and methods for clinical interpretation pending and issued. B.A. McGregor reports grants and personal fees from Exelixis, Pfizer, Calithera, and BMS, and personal fees from SeaGen outside the submitted work. S. Signoretti reports grants from Genentech, Inc. during the conduct of the study; grants from Exelixis, Bristol Myers Squibb, and personal fees from AstraZeneca, CRISPR Theapeutics, Merck, AACR, and NCI outside the submitted work; as well as reports a patent for Biogenex with royalties paid. T.K. Choueiri reports grants, personal fees, and non-financial support from Roche and Genentech during the conduct of the study; and reports institutional and/or personal, paid and/or unpaid support for research, advisory boards, consultancy, and/or honoraria from: Alkermes, AstraZeneca, Aravive, Aveo, Bayer, Bristol Myers-Squibb, Calithera, Circle Pharma, Eisai, EMD Serono, Exelixis, GlaxoSmithKline, Gilead, IQVA, Infinity, Ipsen, Jansen, Kanaph, Lilly, Merck, Nikang, Nuscan, Novartis, Oncohost, Pfizer, Roche, Sanofi/Aventis, Scholar Rock, Surface Oncology, Takeda, Tempest, Up-To-Date, CME events (Peerview, OncLive, MJH and others), outside the submitted work. Institutional patents filed on molecular alterations and immunotherapy response/toxicity, and ctDNA. Equity: Tempest, Pionyr, Osel, Precede Bio, CureResponse. Committees: NCCN, GU Steering Committee, ASCO/ESMO, ACCRU, KidneyCan. Medical writing and editorial assistance support may have been funded by Communications companies in part. No speaker's bureau. Mentored several non-US citizens on research projects with potential funding (in part) from non-US sources/Foreign Components. The institution (Dana-Farber Cancer Institute) may have received additional independent funding of drug companies or/and royalties potentially involved in research around the subject matter. T.K. Choueiri is supported in part by the Dana-Farber/Harvard Cancer Center Kidney SPORE (2P50CA101942-16) and Program 5P30CA006516-56, the Kohlberg Chair at Harvard Medical School and the Trust Family, Michael Brigham, Pan Mass Challenge, Hinda and Arthur Marcus Fund and Loker Pinard Funds for Kidney Cancer Research at DFCI. R.R. McKay reports Consultant/Advisory board for Aveo, AstraZeneca, Bayer, BMS, Calithera, Caris, Dendreon, Exelixis, Eisai, JNJ, Lilly, Merck, Myovant, Novartis, Pfizer, Sanofi, Sorrento Therapeutics, SeaGen, Tempus, and Telix. Institutional research support AstraZeneca, BMS, Exelixis, Oncternal, Tempus. D.A. Braun reports grants from Roche/Genentech during the conduct of the study; personal fees from Fortress Bio Subsidiary, grants and personal fees from Exelixis, grants from AstraZeneca, and personal fees from Targeted Oncology, ASCO Post RCC Roundtable, Abbvie, Aptitude Health, Haymarker, PWW Consulting, Cell Health BioConsulting, OncLive, Catenion, Cancer Network, Aveo, MDedge, Adnovate Strategies, and Cancer Expert Now outside the submitted work. No disclosures were reported by the other authors.

R.M. Saliby: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, project administration. T. El Zarif: Conceptualization, investigation, visualization, methodology, writing–original draft, project administration. Z. Bakouny: Conceptualization, investigation, visualization, methodology, writing–original draft, project administration. V. Shah: Formal analysis, validation. W. Xie: Resources, formal analysis, validation, writing–review and editing. R. Flippot: Formal analysis, investigation, methodology, writing–review and editing. T. Denize: Formal analysis, investigation, methodology, writing–review and editing. M.H. Kane: Resources, writing–original draft, writing–review and editing. K.N. Madsen: Writing–original draft, writing–review and editing. M. Ficial: Resources, methodology, writing–review and editing. L. Hirsch: Resources, writing–review and editing. X.X. Wei: Resources, writing–review and editing. J.A. Steinarter: Resources, writing–review and editing, patient enrollment. L.C. Harshman: Resources, methodology, writing–review and editing. U.N. Vaishampayan: Resources, writing–review and editing. M. Severgnini: Resources, writing–review and editing. D.F. McDermott: Resources, supervision, writing–review and editing. G.-S.M. Lee: Resources, supervision, writing–review and editing. W. Xu: Resources, supervision, writing–review and editing. E.M. Van Allen: Resources, supervision, writing–review and editing. B.A. McGregor: Writing–review and editing, patient enrollment. S. Signoretti: Resources, supervision, methodology. T.K. Choueiri: Resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. R.R. McKay: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. D.A. Braun: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

S. Signoretti and T.K. Choueiri are supported in part by the Dana-Farber/Harvard Cancer Center Kidney SPORE (2P50CA101942–16) and Program 5P30CA006516–56. T.K. Choueiri is supported in part by the Kohlberg Chair at Harvard Medical School and the Trust Family, Michael Brigham, Pan Mass Challenge and Loker Pinard Funds for Kidney Cancer Research at DFCI. Roche/Genentech provided support to this investigator-initiated study (NCT02724878).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.